Maturity of nearby faults influences seismic hazardfrom hydraulic fracturingMaria Kozłowskaa,1, Michael R. Brudzinskia, Paul Fribergb, Robert J. Skoumalc, Nicholas D. Baxtera, and Brian S. Curriea

aDepartment of Geology and Environmental Earth Science, Miami University, Oxford, OH 45056; bInstrumental Software Technologies, Inc., New Paltz, NY12561; and cEarthquake Science Center, US Geological Survey, Menlo Park, CA 94025

Edited by Paul Segall, Stanford University, Stanford, CA, and approved December 21, 2017 (received for review August 29, 2017)

Understanding the causes of human-induced earthquakes isparamount to reducing societal risk. We investigated five casesof seismicity associated with hydraulic fracturing (HF) in Ohio since2013 that, because of their isolation from other injection activities,provide an ideal setting for studying the relations between high-pressure injection and earthquakes. Our analysis revealed twodistinct groups: (i) deeper earthquakes in the Precambrian base-ment, with larger magnitudes (M > 2), b-values < 1, and manypost–shut-in earthquakes, versus (ii) shallower earthquakes in Pa-leozoic rocks ∼400 m below HF, with smaller magnitudes (M < 1),b-values > 1.5, and few post–shut-in earthquakes. Based on geo-logic history, laboratory experiments, and fault modeling, we in-terpret the deep seismicity as slip on more mature faults in oldercrystalline rocks and the shallow seismicity as slip on immaturefaults in younger sedimentary rocks. This suggests that HF induc-ing deeper seismicity may pose higher seismic hazards. Wells in-ducing deeper seismicity produced more water than wells withshallow seismicity, indicating more extensive hydrologic connec-tions outside the target formation, consistent with pore pressurediffusion influencing seismicity. However, for both groups, the2 to 3 h between onset of HF and seismicity is too short for typicalfluid pressure diffusion rates across distances of ∼1 km and arguesfor poroelastic stress transfer also having a primary influenceon seismicity.

induced seismicity | hydraulic fracturing | frequency-magnitudedistribution | b-value | seismic hazard

As the development of unconventional oil and gas resourceshas become more commonplace, the potential for seismicity

induced by wastewater disposal (WD) and hydraulic fracturing(HF) has become an increasingly important issue (1). Recentstudies have shown that the extraordinary increase in the numberof earthquakes with magnitude (M) ≥ 3 in the central andeastern United States over the past decade is largely a result oflarge increases in the amount of wastewater disposal (2–4).Based on these studies, there is a growing tendency to considerWD operations as the primary concern in the assessment ofinduced-seismicity hazards (5, 6). Moreover, these findings havebeen taken as evidence for the physical model where fluid in-jection increases pore fluid pressure, lowering the effective stresson faults and promoting seismic slip. However, a growing num-ber of studies have found evidence that HF alone is also re-sponsible for M ≥ 3 seismicity (7–9) although a given HF well hasshown lower prevalence of inducing M ≥ 3 seismicity than a WDwell (10, 11). While it is plausible that the rapid changes in porefluid pressure are inducing slip on adjacent faults (12), it is un-clear whether appropriate permeability structures exist for rapidtransmission of pore fluid pressures (9). Some recent researchsuggests that poroelastic stress could have a larger influence onseismicity than pore fluid perturbations in cases of both WD andHF (13–16), but others contend that poroelastic stress dissipatestoo quickly from the injection interval (17). However, thesestudies seeking to discern between these mechanisms are eitherstrictly theoretical or target only a single observed sequence.

In this study, we focused on the easternmost part of Ohio,Harrison County, where HF began to target the unconventionalUtica/Point Pleasant tight shale formations in 2012. This areahad essentially no documented seismicity before 2010 (Fig. 1)but has since had over a dozen cases of induced seismicity. Thefirst large sequence occurred in 2011, when WD induced hun-dreds of events M > 1 up to a maximum observed magnitude(MMAX) 4.0 in Youngstown (18, 19). Since then, seismicity as-sociated with WD has also been observed in Trumbull County(MMAX 2.1) (20) and Washington Counties (MMAX 3.1) (10)whereas seismicity associated with HF has occurred in PolandTownship (MMAX 3.0) (9), Belmont and Guernsey Counties(MMAX 2.6) (10), and Harrison County (MMAX 2.6) (21, 22).The seismicity associated with HF has been most frequent inHarrison County, where it was first recorded in October 2013.The relatively slow rate of well drilling and HF stimulationrelative to other shale plays, as well as the lack of nearby WDwells, makes this area an ideal setting for isolating the rela-tionships between high pressure injection and induced seis-micity. To detect and locate additional seismic events inducedby ongoing HF in the area, the Division of Oil and GasResources Management at the Ohio Department of NaturalResources (ODNR) deployed four portable short-period three-component seismometers in Harrison County by November2013. By the end of 2015, another four major sequences ofseismicity were detected using the repeating signal detector(RSD) algorithm (22). The preliminary hypocenters de-termined in that study suggested that the seismicity occurred

Significance

Recent studies have focused on howwastewater disposal wellshave caused dramatic increases in eastern US earthquakes. Wefocused instead on less common cases where hydraulic frac-turing alone has caused earthquakes and found seismicityseparated into two depth zones: a shallow zone on youngerfaults, with more small-magnitude earthquakes than expected,and a deeper zone on older faults, with larger magnitudeearthquakes and seismicity continuing after fracturing stops.Hence, inducing deeper seismicity creates more hazard. Ourobservations are consistent with prior geologic, laboratory,and theoretical work indicating that age and maturity offaults causes the different seismicity patterns. We utilize datafrom well operators to demonstrate that both fluid pressurechanges and rock stress transfer are needed to explain ourobservations.

Author contributions: M.K. and M.R.B. designed research; M.K. and M.R.B. performedresearch; P.F. contributed new reagents/analytic tools; M.K., P.F., R.J.S., and N.D.B. ana-lyzed data; and M.K., M.R.B., and B.S.C. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Published under the PNAS license.1To whom correspondence should be addressed. Email: mkozlow@igf.edu.pl.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1715284115/-/DCSupplemental.

E1720–E1729 | PNAS | Published online February 5, 2018 www.pnas.org/cgi/doi/10.1073/pnas.1715284115

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

7, 2

020

beneath or near the wells that were active at the same time, buta detailed relative location study has not yet been performed.Of particular interest is whether the different wells are stimu-lating seismicity on a larger fault that extends across the entireregion or whether the seismicity occurs on separate smallerfaults. Distinguishing between these scenarios is necessary toestimate the maximum magnitude potential and resulting haz-ard these events pose, and our study is specifically designed toprovide the detailed location analysis for this purpose.Another key component for evaluating the hazard associated

with induced seismicity is to determine the frequency-magnitudedistribution (FMD). The FMD for a given earthquake’s pop-ulation describes the rate of earthquakes across all observedmagnitudes, which is critical for estimating the likelihood oflarger and potentially damaging earthquakes from an initial setof smaller events (23). The Gutenberg and Richter (24) relation(G–R) defines FMD as log10N(M) = a − bM, where N is thecumulative number of earthquakes of magnitude ≥M, and a andb (the so-called b-value) denote constants characterizing anearthquake population’s specifics. As in many self-similar dis-tributions in nature, there is evidence for a long-term constantand universal b-value of 1 (25–27). However, deviations from thisvalue are thought to represent a wide range of physical phe-nomena (28, 29). However, the ultimate importance of charac-terizing b-values is that it is necessary to forecast the likelihoodof larger events. For example, lower b-values indicate more largerevents relative to a typical population and hence a larger hazard.Finally, the depth of fluid injection operations has been suggested

to influence the hazard from induced seismicity (30–32). Based onthe observational correlations and geomechanical and hydrologicmodeling in these studies, the likelihood of induced seismicity ishigher the closer the WD or HF operations are to the Precambriancrystalline basement rocks. The suggested reasoning is that the

faults in these older rocks are more capable of producing seismicityof sufficient magnitudes to be detected (M > 2) and potentiallyhazardous [modified Mercalli intensity scale (MMI) ≥ V].In this study, we performed a detailed investigation of the

largest seismic sequences associated with HF in Harrison County,Ohio, from 2013 to 2015. To better understand the causes of thisseismicity, we utilized improved estimation of event detection,hypocentral locations, frequency-magnitude distributions, andfocal mechanism determinations in conjunction with detailedoperational and production information. We found that theseismicity induced by HF falls into two groups that reveal im-portant differences in fault maturity with depth. These differ-ences influence the seismic productivity both during and afterHF, which indicates that wells inducing deeper events pose sig-nificantly larger hazards. Moreover, the eventual water and hy-drocarbon production of these wells is correlated with the type ofseismicity they produce, providing clues about the extent ofstimulation. Finally, the temporal–spatial relationship betweenHF and seismicity requires triggering mechanisms faster thanpore fluid pressure diffusion.

ResultsSpatial Patterns of Seismicity. Fig. 2 shows our relocated epicen-ters colored according to the associated HF well. The epicentersoutline several east–west-oriented trends, similar to that seen inother cases of induced seismicity in Ohio (10). The maximumhorizontal stress (SHmax) is oriented northeast–southwest ineastern Ohio (33, 34), as demonstrated by well lateral trajecto-ries oriented perpendicular to SHmax to facilitate fracture open-ing (Fig. 2). The apparent east–west fault orientations fromseismicity are ∼30° from SHmax and optimally oriented for shearslip reactivation. The lengths of the bands of seismicity appear tobe ∼1 km or less as the refined relative locations provide evi-dence that seismicity is not occurring on a single through-goingfault. However, there appear to be two cases (Ryser–Davidsonand Tarbert–Hamilton) when HF ∼1 y later induced seismicityon the same fault activated earlier. Fault lengths limited to 1 kmare similar to all previous induced seismicity in Ohio, but thisrepresents the imaging of a multitraced network of faults beinginduced by multiple HF well pads in the United States.

Fig. 1. Map of southwestern Harrison County, Ohio, showing the results fromRSD processing (22). Lines show locations of unconventional well head andtoe, colored by time of HF, restricted to the time range of the RSD analysis.Triangles show seismic stations used in this study. Symbols are also colored bytime and show locations of the strongest events from major (>750 events,circles) and minor (<750 events, diamonds) earthquake sequences determinedby Skoumal et al. (22). Squares mark well heads where stimulation occurredbefore the local network was installed, but HF correlates temporally withseismic activity and S-wave minus P-wave (S − P) times recorded by O53A areconsistent with earthquakes occurring near these wells. (Inset) Map of Ohioshowing earthquakes before 2010 (red), HF wells (yellow), and induced seis-micity (blue) (20, 30). Square marks location of larger map.

Fig. 2. Map showing five major sequences in Harrison County from 2013 to2015 analyzed in this study. Curved lines indicate horizontal well drill paths.Stimulation stages (squares) and double-difference relocated seismicity(circles) are shaded using a time scale since the start of HF, with a differentcolor for each well pad. Circle diameter shows mean horizontal locationuncertainty. Nearest seismic station is shown as a triangle.

Kozłowska et al. PNAS | Published online February 5, 2018 | E1721

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

7, 2

020

Another key feature is that seismicity tends to occur within orvery near the footprint of the well pad to which the seismicity istemporally correlated (Fig. 2). The footprint size is based on theextent of varying numbers of horizontal well laterals (1H to 10H).

Fig. 3 shows the calculated 3D separations between seismicity andthe most recent HF stages (squares, Fig. 2). These values vary from140 to 1,220 m, similar to the Poland Township case, where onlycertain stages within 800 m of the fault induced seismicity (9).

Fig. 3. Temporal distribution of seismicity detected using RSD, showing magnitudes of seismic events (black) together with HF stages labeled with individualwell names (bottom blue bars), shown for all studied wells (A–D). Top colored circles show 3D distance from each located earthquake to nearest HF stage thatoccurred before the earthquake. Well locations and colors correspond to those in Fig. 2.

E1722 | www.pnas.org/cgi/doi/10.1073/pnas.1715284115 Kozłowska et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

7, 2

020

Cross-sections through the double difference relocationsrevealed a key finding that seismicity is separated into two dif-ferent depths (Fig. 4). The previous work in Youngstown, PolandTownship, and Trumbull County (19, 9, 20), as well as work onthe Ryser pad seismicity in Harrison County (21), found thatinduced seismicity occurred in the crystalline Precambrian rocksof the uppermost basement. Our analysis found that nearly all ofthe locatable Ryser and Hamilton pad seismicity occurred in theuppermost basement (Fig. 4), consistent with the earlier findings.The basement depth is well-determined in Ohio based on anintegrated study of multiple data sources (35). We also foundthat over half of the Davidson seismicity occurred in the up-permost basement, very near the Ryser seismicity (Fig. 4).However, the remaining Davidson seismicity occurred signifi-cantly above the basement, closer to but still below the targetUtica–Point Pleasant formations. Seismicity associated with theVozar well pad also occurred at shallower depths. Nearly all ofthe Tarbert seismicity was also at similar shallow depths, exceptfor the two notably larger (M ∼ 2) events. Utilizing the clue thatthe two larger Tarbert events occurred deeper than the otherTarbert events, we also noted that the deeper seismicity associ-ated with the Ryser and Hamilton pads included more largeevents than the shallower seismicity associated with the David-son and Vozar pads.

Temporal Patterns of Seismicity. Fig. 3 shows the timing of HFstages for the five multiple-lateral well pads associated with thelargest seismic sequences. The onset of seismicity was well cor-related with the onset of HF at these wells. Not all well pads havedetailed (minute precision), publically available stimulation re-ports, but there are two well pads with this level of detail and noother well pads with HF in close proximity in the days before theonset of HF. For the Ryser well pad, there were only 120 minfrom the initiation of HF to the onset of seismicity, and for theDavidson well pad the time offset was 171 min. For comparison,the Poland Township case a few counties away only had seis-micity during a subset of stages (9), but it appears the time offsetwas 109 min based on the correlation between HF and seismicity.In general, seismicity was well-correlated with the onset of HF

for all well pads, and there were distinct bursts of seismicity thatcorrelated with individual stages of HF (Fig. 3). However, theseismicity only halted at shut-in for the Tarbert and Vozar wellpads. Seismicity continued in the same general locations for overa month after the Ryser and Hamilton pads shut-in although the

operations at the nearby Pettay pad following the Hamilton padcreated some level of uncertainty in that case. This seismicitythat continues after shut-in occurred at lower seismicity ratesthan during HF and decayed over time (Supporting Information),but the magnitude ranges were similar to that during HF. Forthis reason, we believe that there is a distinctly different patternof seismicity following the Tarbert and Vozar HF and the lack ofseismicity after shut-in of those well pads was not due to lack ofdetection. It is important to note that the occurrence of seis-micity did not influence the HF operations except for Hamilton,where communication between the operator and regulatorresulted in stimulation being halted on well laterals 6H and 8H(Fig. 2) for over 2 wk following the strongest earthquake and theinjection pressures and volumes were decreased afterward.

Frequency-Magnitude Distribution. To better understand the mag-nitude differences between the two depth groups, we estimatedFMDs for each sequence and compared them with typical G–Rrelations (Table S1). The FMDs differed for the two separatedepth zones: the deeper sequences associated with the Hamiltonand Ryser wells had overall low b-values (0.72 and 0.76) and theshallower sequences from the Tarbert and Vozar wells had highvalues (1.54 and 1.91). The Davidson well sequence, which con-sisted of two clusters differing in depths, showed a b-value (1.3) inbetween that of the two clusters. Similar large variation of b-valueswere reported for different cases of HF-induced seismicity, typ-ically ranging from 0.8 to 1.0 for out-of-zone seismicity thatreaches larger magnitudes (21, 9) to 2.0 for smaller near-targetmicroseismic events (36–38). Comparing b-values betweendifferent studies can be difficult as authors often use differenttypes of magnitudes and different techniques for magnitude ofcompleteness (Mc) estimation, both influencing the b-value.Since our study area was relatively small and we detected allevents using the same station and the same approach, it is le-gitimate to compare the obtained b-values and to associatetheir variation with geological or operational factors.To look for any temporal changes of b-value like those seen in

the study of WD in Youngstown (19), Fig. 5 separates the cat-alogs into an early, late, and post-HF phases (if it existed). Forall sequences except Tarbert, the b-value was highest in the earlyphase. For the deeper Ryser and Hamilton seismicity, theb-value was slightly higher (∼0.9) initially, but decreased duringstimulation to values (∼0.7) that are similar to those of seismicityinduced by WD in Ohio (10, 19, 20). Authors analyzing b-value

Fig. 4. Cross-sections taken (A) east–west and (B) north–south through the relocated seismicity shown in Fig. 2. Colors correspond to those in Fig. 2. Circlediameter shows mean vertical location uncertainty in our study. Curved lines are well lateral drill paths. Dashed line marks the top of the Precambrianbasement rocks (35).

Kozłowska et al. PNAS | Published online February 5, 2018 | E1723

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

7, 2

020

variations in areas where underground fluid injection is per-formed associate changes in the parameter with changes ofpressure gradient (29, 39). While most wells showed a gradualdecrease over time, the Davidson case showed a larger changethat corresponds to a change in depth. The early phase hadshallower seismicity and high b-values (1.65) while the late phasehad deeper seismicity and lower b-values (1.12). Closer exami-

nation of the Tarbert exception reveals that the late phase FMDdid not follow a log–log power law relationship due to a fewmuch larger events (Figs. 3 and 5). The late phase for Vozarshowed this pattern to a lesser degree, and a similar trend wasobserved for the Trumbull and Youngstown WD wells (20).The FMDs of several phases of seismicity appeared to be

truncated with a curvature that produces fewer larger eventsrelative to what is expected for a linear G–R (power law) FMDsimilar to that of Huang and Beroza (40). For each phase ofseismicity, we used an F-ratio test to determine the confidencelevel associated with improved fit of a truncated form of the G–Rcurve (41, 42) over the linear G–R, taking into account the extradegree of freedom. For later phases of Tarbert and Vozar,FMDs did not follow a linear G–R or truncated G–R, but theyappear to be similar to a nonlinear trend observed for theTrumbull and Youngstown WD wells (10). For Davidson Lateand Hamilton Post, the linear G–R fit was better than thetruncated G–R. For Vozar Early, truncated G–R was a better fitat 98% confidence. For all other cases, truncated G–R was asuperior fit at >99.999% confidence.

Focal Mechanism and Waveforms Similarity Analysis. Due to limitednumber of local stations, we were only able to calculate a focalmechanism for the two strongest (M > 2.5) events from Ryserand Hamilton sequences that were well-recorded on regionalstations; their focal mechanisms are presented in Friberg et al.(21) and Fig. S2, respectively. Both focal mechanisms are similarand consistent with those determined for strongest events inYoungstown (18) and Poland Township (9), showing similareast–west faults orientation as the hypocentral distribution. Allmentioned events were located within the Precambrian base-ment, suggesting that a fault network with parallel traces that isoptimally oriented exists across eastern Ohio.Unfortunately, none of the shallower earthquakes were large

enough to determine a focal mechanism. Their proximity to thetarget formation and high b-values might suggest they occurredvia a similar process to that of operationally induced microseis-micity, which is thought to represent creation of new, or openingof preexisting, fractures and hence have a non–double-couplefocal mechanism representing their tensile character (43). How-ever, the detection tool used to create the events we analyzed wasbased on high waveform similarity, which should retrieve eventswith similar focal mechanisms. Visual comparison of first-motionpolarities for deep and shallow events confirmed the similarity(Fig. S3). HF-generated fracture orientations are usually differ-ent from preexisting fault orientations (44). Moreover, Wolhartet al. (44) illustrates how typical shale fracturing events occur atM < −2 and that larger events represent reactivation of slip onpreexisting faults outside the target formation. Together, thesefindings suggest that the shallow seismicity that we observedrepresents on-fault slip rather than a fracturing process and thatboth Precambrian and Paleozoic faults experience a commonfaulting regime.

Operational Data from HF Wells. To look for any patterns in in-jection parameters that correlate with the depth and b-valuepatterns we observed, we used individual stage values listed instimulation reports to calculate mean (Table S2) and maximum(Table S3) injection pressures and volumes. We also examinedvalues from three neighboring well pads (Pettay, Gustina, andPuskarich) (Fig. 1) with HF in our study period but with onlyvery weak seismicity (22). The injection parameters did notcorrelate with the depth and FMD patterns we observed. Ryserand Hamilton had slightly higher breakdown pressures thanVozar and Tarbert (Tables S2 and S3), but these variations aresimilar to that observed at wells with only weak seismicity, in-dicating no clear relationship between injection parametersand seismicity.

Fig. 5. Frequency-magnitude distributions of five detected seismic se-quences, divided into HF phases (A–L). N/Ntot is the number of events at orabove a given magnitude divided by the total number of events. Colorsfollow color scale in Fig. 2. The black circles are below the estimated Mc. b-values were calculated using maximum-likelihood estimate, and uncer-tainties were estimated using bootstrap resampling. The black line repre-sents maximum-likelihood G–R fit, and red curves represent truncated G–Rfit. Please note that Tarber and Vozar Late Frac seismicity (B and D) do notobey log–log power law distribution.

E1724 | www.pnas.org/cgi/doi/10.1073/pnas.1715284115 Kozłowska et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

7, 2

020

We also analyzed water and gas production data reported toODNR quarterly for the five wells with seismicity to look forother clues about what led to the two groups of seismicity (Fig.6). For comparison, we again included data from wells with littleseismic activity: Pettay, Puskarich, and Gustina. The wells withdeep seismicity (blue, green, and yellow curves in Fig. 6) showedhigher values of water production over the first few quarters thanthe wells with shallow (purple and red) or no seismicity (dashedgray). Water production typically fell off over time (Fig. 6A), butthe Ryser well lateral closest to the Davidson wells (3H) saw anunexpected increase during HF of the Davidson wells (greenlines). This is one of the few times when seismicity during HF ofone well pad aligned with seismicity during HF of a later well pad(Fig. 2). The other time seismicity appeared to align was for theTarbert and Hamilton wells, and a similar increase in waterproduction occurred for the nearest Tarbert well lateral (5H)during HF of the Hamilton wells (e.g., 8H and 2H). Waterproduction is often correlated with gas production, but the wellswith deep seismicity did not have larger gas production (Fig. S4)so it cannot explain the patterns. For example, the Ryser wellhad one of the largest initial gas production values, but it wassimilar to two wells with weak seismicity. However, if we plot thegas production aligned on the start time with values normalized(Fig. 6B), wells with deep seismicity have production values thatfall off faster (lower normalized values at later times, particularly

after 1 y) than the wells with shallow or no seismicity. We hy-pothesize that the fracturing and propping process appears to beless effective for wells that induce deep seismicity due to moreextensive hydrologic connections outside the target reservoir.

DiscussionOur observations indicate that there is a network of preexisting,critically stressed faults in this area before HF, primarily basedon the rapid onset of optimally oriented seismicity once HFbegins. The relocated seismicity indicates that preexisting faultsoccur both in the Precambrian and the Paleozoic section. Wehypothesize that both the Paleozoic and Precambrian seismicityoccur on left-lateral strike-slip faults that are optimally orientedrelative to SHmax. While the hypocenters do not illustrate a fullconnection between the two depth zones, we hypothesize aconnection via linked, high-angle fault systems that originate inthe master fault in the crystalline basement and extend upwardthrough or near to the target Utica–Point Pleasant reservoirinterval. Strike-slip systems experiencing convergence, as ap-pears to be the case based on focal mechanisms and proximity tothe Appalachian orogenic belt, are expected to develop a posi-tive flower structure (45). Evidence for this geometry was re-cently found ∼100 km south in Washington County throughcombined analysis of seismicity and mapping the subsurfacestructure in a case of WD-induced seismicity (46).Fig. 7 shows seismic rupture areas as circles synthetically

computed from the average FMDs for the Paleozoic and Pre-cambrian seismicity groups to visually illustrate the differences inb-value and maximum magnitudes. The high b-value of theshallow seismicity (1.75) results from so many small events thatthe fault patch appears to be full of similar sized seismicitywhereas the low b-value of the deep seismicity (0.75) is caused by

Fig. 6. Quarterly production data from five wells with seismicity we ana-lyzed (colored lines). Gray lines show data from three other well pads withHF during our study with only weak seismicity. (A) Water production. (B)Natural gas production aligned on the first production quarter, normalizedto the largest value. Error bars show the SD for all 269 wells in HarrisonCounty.

Fig. 7. Schematic cross-section illustrating differences between the shallowseismicity in the Paleozoic sedimentary rocks and the deep seismicity in thePrecambrian basement rocks. Circles show seismic rupture areas syntheticallycomputed from the average FMDs, utilizing a magnitude to circular slip areaapproximation (81), but with a low stress drop [0.1 MPa (megapascals)] toproduce visible circle sizes. Circle locations are randomly assigned in a squarearea in the depth range observed. The minimum size plotted (M > −1) isfrom our average Mc. Strata are colored by predominant rock type, withdepths estimated from the closest deep well logs. Curved lines show examplewell paths.

Kozłowska et al. PNAS | Published online February 5, 2018 | E1725

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

7, 2

020

more magnitude variability and larger events visually dominatethe fault patch. Efforts to understand high b-values of opera-tionally induced microseismicity recently led to the idea thatmechanical bed thickness in the target formation limits thelength of fractures (47, 48). However, waveform and first motionsimilarity between Paleozoic and Precambrian seismicity indi-cates that the high b-values of our Paleozoic seismicity are notassociated with the fracturing process (Fig. S3). Moreover, thisseismicity occurs ∼400 m below the target formation, straddlingseveral different types of strata (Figs. 4 and 7), with magnitudeslarger than any operationally induced microseismicity (49).We interpret that the difference in b-values for shallow versus

deep seismicity is due to differences in fault maturity based onthe different age of the rocks that host the faults (∼450 My forPaleozoic; ∼1 Gy for Precambrian). Recent laboratory stick-slipexperiments revealed that the acoustic emission events producedon rough, fractured surfaces have much higher b-value (1.2) thanthose on smooth, polished surfaces (0.7) (50). These resultssuggest that more mature faults develop smoother surfaces withbroader spatial areas of homogeneous stress that promote largerruptures, leading to lower b-values, whereas younger faults areassociated with rougher surfaces and stress field heterogeneitiesthat have narrower spatial extent, leading to smaller rupturesizes and higher b-values. Studies have also found that the size ofseismic asperities increases with increasing fault maturity (51), asmore slip on more mature faults appears to reduce the faultroughness (52). Laboratory studies suggest that increasing strainlocalization is an important part of this process (53). Depthdependence of b-values was also observed for seismicity inCalifornia (54), where the changes were attributed to greaterheterogeneity of shallow formations and smaller lithostatic stressin comparison with deeper formations. However, that study an-alyzed a much greater depth range (15 km) than the one in thisstudy (1.5 km), such that stress variations are unlikely to be re-sponsible for the vastly different b-values in our study. We cannotrule out that material property changes between the crystallinebasement and sedimentary rocks are contributing to the differ-ences in seismicity, but we find no studies proposing or doc-umenting different b-values due to material properties. Thematurity difference hypothesis is also justified from a geologicalpoint of view.Subsurface mapping of Paleozoic stratigraphic units in con-

junction with limited seismic reflection data indicate thatbasement-involved faults in eastern Ohio have experiencedmultiple periods of deformation throughout the Paleozoic. InHarrison County, both NW–SE and E–W trending fault zonesdisplay Cambrian-aged extensional deformation, followed byPennsylvanian–Permian transpressional reactivation (55). Throughoutsoutheastern Ohio, similar faults are associated with variablethickness trends in Cambrian, Silurian, Devonian, and Pennsyl-vanian strata, suggesting that multiple episodes of reactivationthroughout the Paleozoic developed along zones of weakness inthe Precambrian crystalline basement (55–59). Given this his-tory, associated fault zones would be significantly more maturelower in the stratigraphic section, with the highest degree ofmaturity in the Precambrian crystalline basement. It is also im-portant to note that these faults are observed extending from thePrecambrian basement into the overlying Paleozoic section,commonly tipping out in Upper Ordovician–Devonian strata (55,57–59). As such, these faults and associated fracture zones mayprovide permeability pathways from hydraulic-fracturing tar-gets in the Ordovician Utica–Point Pleasant reservoir into theLower Paleozoic sedimentary strata and underlying Precambriancrystalline basement.The seismicity we detected occurred rapidly after the onset of

HF (120–171 min). While we don’t know the exact location ofthe first seismic event in the Ryser, since the local network wasinstalled several weeks later, the O53A waveform similarity in-

dicates that it occurred in the same group as those located withthe network. Based on this, there is ∼1,200 m between the firststage at the Ryser well and the first earthquake. We can be moreprecise for Davidson, where the first stage is separated by 820 mfrom the first earthquake. For a saturated homogeneous me-dium, these values correspond to hydraulic diffusivities of 18 m2/sand 5.2 m2/s, respectively, which are more than an order ofmagnitude larger than other estimates of diffusion-triggeredseismicity in shallow fault zones and induced seismicity (60,61). This observation is more consistent with recent research thathas suggested that poroelastic stress could have a larger influ-ence on seismicity than pore fluid pressure changes in cases ofboth WD and HF (13–15). In particular, Deng et al. (14), dis-cussing a case of HF in Canada, found that the effect ofporoelastic stress transfer was large enough to induce seismicityeven at distances >1 km from a single well. In our case, thismechanism appears to play a role throughout HF as many dif-ferent stages along the horizontal well that extend over a kilo-meter continue to trigger seismicity on the same fault. Thiswould be difficult to achieve with pore fluid pressure changesalone as each spatially separated stage would need to create fluidpathways that connect to the same fault. Moreover, we do notsee an increasing trend in distance from the stage to the seis-micity over time (Fig. 3) that would otherwise be expected if porepressure diffusion were the sole cause. In addition, the strongtemporal variations in injection associated with HF are morelikely to cause poroelastic stress changes (13, 16).While Deng et al. (14) argued that poroelastic stresses are the

dominant factor responsible for a sequence of earthquakes M ≥2 during HF, most studies have argued that pore fluid pressurechanges are the dominant factor inducing seismic events in casesof HF (39, 62, 63), and our study finds evidence for this processas well. Foremost, water production increases in neighboringwells when induced seismicity occurs (Fig. 6A) while we do notobserve a similar jump in gas production (Fig. S4). If poroelasticstress was increasing pressure near the well, it should increaseboth water and gas production. Instead, the increased waterproduction indicates more hydraulic conductivity that shouldenable pore fluid pressure changes outside the shale target.Likewise, the decrease in b-values over the course of HF forwells that induced deep seismicity (Fig. 5 G, H, J, and K) suggestthat pore fluid pressures may increase in these fault zones duringthe HF. This is based on studies showing that increased porefluid pressure, caused either by natural processes in the sub-duction zone or by high pressure fluid injection, results in low-ered effective normal stress and decreased seismicity b-value (29,64–66).The truncation of the FMD during HF (Fig. 5 A, C, E, G, H,

and J) also indicates the effect of pore fluid pressure on seis-micity as injection progresses. A truncated FMD was found for adetailed study of wastewater disposal-induced seismicity alongthe Guy–Greenbrier Fault in Arkansas (40). In that study, thesum of injection wastewater volumes provided a good constrainton the seismic moment using the relationship of McGarr (67),consistent with the idea that the finite size of the mediumstimulated by fluid injection limited earthquake source volumes(68). So, while most studies have concluded that either pore fluidpressure changes or poroelastic stress transfer is the dominantfactor for inducing seismicity, we believe our study demonstratesthat both mechanisms are necessary to explain the observations.Intriguingly, the Ryser and Hamilton cases with post–shut-inseismicity show the b-value rebounding and the FMD movingfrom truncated G–R to linear G–R during the post-HF timeperiod (Fig. 5).This matches the pattern seen by Huang and Beroza (40) after

the termination of wastewater disposal injections, where seis-micity continued even after the earthquakes stopped migrat-ing along the Guy–Greenbrier Fault. This observation matches

E1726 | www.pnas.org/cgi/doi/10.1073/pnas.1715284115 Kozłowska et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

7, 2

020

conclusions made by Segall and Lu (13) that postinjectionearthquakes are triggered by continuously increased Coulombstress away from the injection point. The process of stress re-laxation, featured by diffused and persistent seismicity, followsDieterich’s seismicity rate model (69). The absence of post–shut-in seismicity for the Paleozoic seismicity in our study is similar toobservational studies of aftershocks along shallow faults thatfound sedimentary rocks do not host aftershocks at magnitudeswe can record (70, 71). The limited aftershocks were interpretedto be due to different rheological and frictional behavior ofsedimentary rocks relative to metamorphic or igneous rocks.Subsequent modeling of seismogenic faults with damage rheol-ogy found that lower effective viscosity of sedimentary rockscould reproduce the paucity of aftershocks (72). Laboratorymeasurements of shales and sands have shown time-dependentviscous behavior such that differential stress changes would bedissipated over time scales on the order of days (73, 74). Thissupports the notion that the differential stress caused by inducedearthquakes is relaxed rapidly in the Paleozoic section due to thelower effective viscosity of sedimentary rocks.Our results suggest that the seismic hazards associated with

HF are largest when Precambrian faults are being activated. Thematurity of these faults appears to allow larger ruptures and agreater ratio of larger to smaller magnitudes. Seismicity alsoextends longer after injection, apparently due to slower stressrelaxation in basement rocks. In contrast, less mature faults andweaker rock layers in the Paleozoic appear to limit magnitudesand post–shut-in activity. Seismicity in our study area has not yetreached M3, but this may be fortuitous considering the totalvolume injected at the well pads ranges from 0.48 to 1.74 × 106

barrels (bbl), with the largest being 3.5 times the volume injectedat the Youngstown WD well that produced an M4.0 earthquake.The relationship between injected volume and moment magni-tude of McGarr (67) indicates that the maximum magnitude forthese well pads would be moment magnitude (Mw)4.2 to Mw4.5.There have been several M > 4 earthquakes induced by HF inremote Alberta, Canada, with local magnitude (ML)4.4 toML4.8 and Mw3.9 to Mw4.1 (11, 62), so there does not appear tobe a limitation based on the HF process. Detailed analysis of thefirst M ≈ 4 case in Alberta found that the largest event occurredin the basement, ∼700 m below the target formation (63). Arecent ML3.7 earthquake on June 3, 2017, occurred ∼30 kmsouth of our study region in proximity to active HF and waspreliminarily located in the Precambrian (geosurvey.ohiodnr.gov/quakes-2010-to-present-pgs/batesville-june-03-2017), which sup-ports the notion that HF can induced M ≈ 4 earthquakes inEastern Ohio, raising the likelihood of a damaging event.

Materials and MethodsSeismic Data. All analyzed earthquakes were detected using the RSD algo-rithm (22) using an excellent EarthScope station (O53A). Relocations of thestrongest of detected events are possible due to the deployment of localstations. Four short-period, three-component sensors were deployed byODNR (OHH1, OHH2, OHH3, and OHH4) in October 2013, but two of themwere moved to other areas of Ohio in May 2014. In September 2015, a

broadband seismometer was deployed further east to enclose the seismicityby Miami University (MUH1) (Fig. 1).

Hypocentral Location.Our location analysis followed the approach of previouswork in Youngstown (19), Poland Township (9), and Trumbull County (20),Ohio using elocate (75) to establish a catalog of location for all matchedevents and hypodd (76) to refine the relative locations (see Supporting In-formation for more detailed discussion of the procedure and resulting un-certainties). We have tested multiple velocity models to specifically verify thedepth distribution of analyzed events, but having at least one station withinan epicentral distance less than the focal depth from all analyzed sequencesplays a major role in constraining the hypocentral depths with lowuncertainty (77).

Earthquake FMD. The initial earthquake catalog of origin times and Richtermagnitudes is from the enhanced repeating seismicity detector of Skoumalet al. (22). We focused on the four major seismicity sequences with at least700 events (Fig. 3), but one case consisted of two spatially separate clustersassociated with the Davidson and Vozar wells. We calculated FMDs of thesefive seismic sequences (Fig. 5), following the Gutenberg–Richter b-value es-timation of Bender (78):

b= 1��lnð10Þ× �

Mavg-ðMC-Mbin=2Þ��

where Mavg is the sampling average of the magnitudes, MC is the magnitudeof completeness, and Mbin is the bin size of earthquake magnitudes. MC wasdetermined using the maximum curvature algorithm (79). The results foreach sequence are listed in Table S1. Uncertainties were estimated usingbootstrap resampling (Fig. 5 and Table S1). The mean difference from un-certainties calculated with the method of Aki (42) is <0.01. To calculate thetruncated G–R fit, we used the equation of Caputo (41). We then used anF-ratio test to determine the confidence level associated with improved fit ofthe truncated G–R curve over the linear G–R. Fit was estimated by calculatingthe χ2 statistic for each model misfit to the FMD observations above the MC.

Focal Mechanism. We calculated a fault-plane solution for the largest event,which was just large enough to make reliable identifications of first-motionpolarities at local and nearby regional stations. We used FocMec to perform agrid search of the focal sphere for a double-couple solution (80) and thentook the median solution with zero polarity errors. A grid-search fit is jus-tified by previous nearby studies that found clear double-couple mechanismsalong apparently preexisting basement faults (9, 10, 18, 21).

HF Well Data. Stimulation reports, drill surveys, and production data wereretrieved from theODNRonlinedatabase (https://gis.ohiodnr.gov/MapViewer/?config=oilgaswells). Gas and water production were reported each quarter(Fig. 6), such that wells that begin flowback near the end of a quarter will havelow values in their first quarter despite daily flowback being largest initiallyand decreasing over time.

ACKNOWLEDGMENTS. This research benefited from discussions withD. Eaton, R. Schultz, J. Shemeta, and T. Tyrrell. Seismic data were obtainedfrom the Incorporated Research Institutions for Seismology (IRIS) Data Manage-ment Center (www.iris.edu) and ODNR. Plots were made using Generic Map-ping Tools v.4.2.1 (www.soest.hawaii.edu/gmt). Support was provided byNSF Grant 1614942, US Geological Survey National Earthquake Hazards Re-duction Program Grant G15AP00089, a Kosciuszko Foundation Grant foryear 2016/2017, and the ODNR Geologic Survey. Any use of trade, firm, orproduct names is for descriptive purposes only and does not imply endorse-ment by the US Government.

1. National Academy of Sciences (NAS) (2012) Induced Seismicity Potential in Energy

Technologies (Natl Acad Press, Washington, DC).2. Ellsworth WL (2013) Injection-induced earthquakes. Science 341:1225942.3. McGarr A, et al. (2015) Geophysics. Coping with earthquakes induced by fluid in-

jection. Science 347:830–831.4. Weingarten M, Ge S, Godt JW, Bekins BA, Rubinstein JL (2015) INDUCED SEISMICITY.

High-rate injection is associated with the increase in U.S. mid-continent seismicity.

Science 348:1336–1340.5. Petersen MD, et al. (2016) 2016 One-Year Seismic Hazard Forecast for the Central and

Eastern United States from Induced and Natural Earthquakes (US Geological Survey,

Denver, CO), Open-File Report 2016–1035.6. Rubinstein JL, Mahani AB (2015) Myths and facts on wastewater injection, hydraulic

fracturing, enhanced oil recovery, and induced seismicity. Seismol Res Lett 86:

1060–1067.

7. Holland A (2013) Earthquakes triggered by hydraulic fracturing in south-central

Oklahoma. Bull Seismol Soc Am 103:1784–1792.8. Clarke H, Eisner L, Styles P, Turner P (2014) Felt seismicity associated with shale gas

hydraulic fracturing: The first documented example in Europe. Geophys Res Lett 41:

8308–8314.9. Skoumal RJ, Brudzinski MR, Currie BS (2015a) Induced earthquakes during hydraulic

fracturing in Poland Township, Ohio. Bull Seismol Soc Am 105:189–197.10. Skoumal RJ, Brudzinski MR, Currie BS (2015c) Distinguishing induced seismicity from

natural seismicity in Ohio: Demonstrating the utility of waveform template matching.

J Geophys Res Solid Earth 120:6284–6296.11. Atkinson GM, et al. (2016) Hydraulic fracturing and seismicity in the western Canada

sedimentary basin. Seismol Res Lett 87:631–647.12. Shapiro SA, Dinske C (2009) Fluid-induced seismicity: Pressure diffusion and hydraulic

fracturing. Geophys Prospect 57:301–310.

Kozłowska et al. PNAS | Published online February 5, 2018 | E1727

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

7, 2

020

13. Segall P, Lu S (2015) Injection‐induced seismicity: Poroelastic and earthquake nucle-ation effects. J Geophys Res Solid Earth 120:5082–5103.

14. Deng K, Liu Y, Harrington RM (2016) Poroelastic stress triggering of the December2013 Crooked Lake, Alberta, induced seismicity sequence. Geophys Res Lett 43:8482–8491.

15. Goebel TH, Weingarten M, Chen X, Haffener J, Brodsky EE (2017) The 2016 Mw5.1Fareview, Oklahoma earthquakes: Evidence for long-range poroelastic triggeringat >40 km from fluid disposal wells. Earth Planet Sci Lett 472:50–61.

16. Barbour AJ, Norbeck JH, Rubinstein JL (2017) The effects of varying injection rates inOsage County, Oklahoma, on the 2016 Mw 5.8 Pawnee earthquake. Seismol Res Lett88:1040–1053.

17. Lele SP, Tyrrell T, Dasari GR, Symington WA (2017) Geomechanical analysis of hydraulicfracturing induced seismicity at Duvernay field in western Canadian sedimentary basin.GeoConvention 2017. Available at https://www.geoconvention.com/archives/2017/161_GC2017_Geomechanical_Analysis_of_Hydraulic_Fracturing_Induced_Seismicity.pdf.Accessed August 30, 2017.

18. Kim W-Y (2013) Induced seismicity associated with fluid injection into a deep well inYoungstown, Ohio. J Geophys Res Solid Earth 118:3506–3518.

19. Skoumal RJ, Brudzinski MR, Currie BS, Levy J (2014) Optimizing multi-station earth-quake template matching through re-examination of the Youngstown, Ohio se-quence. Earth Planet Sci Lett 405:274–280.

20. Skoumal RJ, Brudzinski MR, Currie BS (2015) Microseismicity induced by deepwastewater injection in Southern Trumbull County, Ohio. Seismol Res Lett 86:1326–1334.

21. Friberg PA, Besana-Ostman GM, Dricker I (2014) Characterization of an earthquakesequence triggered by hydraulic fracturing in Harrison County, Ohio. Seismol Res Lett85:1295–1307.

22. Skoumal RJ, Brudzinski MR, Currie BS (2016) An efficient repeating signal detector toinvestigate earthquake swarms. J Geophys Res Solid Earth 121:5880–5897.

23. Weichert DH (1980) Estimation of the earthquake recurrence parameters for unequalobservation periods for different magnitudes. Bull Seismol Soc Am 70:1337–1346.

24. Gutenberg B, Richter CF (1944) Frequency of earthquakes in California. Bull SeismolSoc Am 34:185–188.

25. Frohlich C, Davis S (1993) Teleseismic b-values: Or, much ado about 1.0. J Geophys Res98:631–644.

26. Kagan YY (1999) Universality of the seismic moment-frequency relation. Pure ApplGeophys 155:537–574.

27. Wech A, Kenneth W, Creager C, Houston H, Vidale J (2010) An earthquake-likemagnitude–frequency distribution of slow slip in northern Cascadia. Geophys ResLett 37:L22310.

28. Schorlemmer D, Wiemer S, Wyss M (2005) Variations in earthquake-size distributionacross different stress regimes. Nature 437:539–542.

29. Bachmann CE, Wiemer S, Goertz-Allmann BP, Woessner J (2012) Influence of pore-pressure on the event-size distribution of induced earthquakes. Geophys Res Lett 39:L09302.

30. Skoumal RJ, Brudzinski MR, Currie BS (2016) Proximity of Precambrian basement af-fects the likelihood of induced seismicity in the Appalachian, Illinois, and WillistonBasins. 2016 Southern California Earthquake Center Annual Meeting. Available athttps://www.scec.org/publication/6570. Accessed October 16, 2017.

31. Zhang Y, et al. (2013) Hydrogeologic controls on induced seismicity in crystallinebasement rocks due to fluid injection into basal reservoirs. Groundwater 51:525–538.

32. Morris AP, et al. (2017) Lessons learned from the Youngstown, Ohio induced earth-quake sequence January 2011 to January 2012. J Rock Mech Geotech Eng 9:783–796.

33. Zoback ML (1992) Stress field constraints on intraplate seismicity in eastern northAmerica. J Geophys Res Solid Earth 97:11761–11782.

34. Hurd O, Zoback MD (2012) Intraplate earthquakes, regional stress and faultmechanics in the central and eastern US and southeastern Canada. Tectonophysics581:182–192.

35. Baranoski MT (2013) Structure contour map on the Precambrian unconformity surfacein Ohio and related basement features (Ohio Department of Natural Resources, Di-vision of Geological Survey, Columbus, OH), Map PG-23, Version 2.0.

36. Maxwell SC, et al. (2009) Fault activation during hydraulic fracturing. SEG TechnicalProgram Expanded Abstracts 2009 (Society of Exploration Geophysicists, Tulsa, OK),pp 1552–1556.

37. Downie RC, Kronenberger E, Maxwell SC (2010) Using microseismic source parametersto evaluate the influence of faults on fracture treatments: A geophysical approach tointerpretation. SPE Annual Technical Conference and Exhibition (Society of PetroleumEngineers, Richardson, TX).

38. Zhou R, Huang GD, Snelling P, Thornton MP, Mueller M (2013) Magnitude calibrationfor microseismic events from hydraulic fracture monitoring. SEG Technical ProgramExpanded Abstracts 1013 (Society of Exploration Geophysicists, Tulsa, OK), pp2145–2149.

39. Mousavi SM, Ogwari PO, Horton SP, Langston CA (2017) Spatio-temporal evolution offrequency-magnitude distribution and seismogenic index during initiation of inducedseismicity at Guy-Greenbrier, Arkansas. Phys Earth Planet Inter 267:53–66.

40. Huang Y, Beroza GC (2015) Temporal variation in the magnitude-frequency distri-bution during the Guy-Greenbrier earthquake sequence. Geophys Res Lett 42:6639–6646.

41. Caputo M (1976) Model and observed seismicity represented in a two-dimensionalspace. Ann Geofis 29:277–288.

42. Aki K (1965) Maximum likelihood estimate of b in the formula log N = a - bM and itsconfidence limits. Bull Earthquake Res Inst Univ Tokyo 43:237–239.

43. �Sílený J, Hill DP, Eisner L, Cornet FH (2009) Non–double-couple mechanisms of mi-croearthquakes induced by hydraulic fracturing. J Geophys Res Solid Earth 114:B08307.

44. Wolhart SL, et al. (2006) Hydraulic fracture diagnostics used to optimize developmentin the Jonah field. SPE Annual Technical Conference and Exhibition (Society of Pe-troleum Engineers, Richardson, TX).

45. Harding TP (1985) Seismic characteristics and identification of negative flowerstructures, positive flower structures, and positive structural inversion. AAPG Bull 69:582–600.

46. Free JC (2017) Relationships between pre-existing structure, regional stress orienta-tion, & seismicity induced by wastewater injection, Washington County, Ohio. BSthesis (Miami University, Oxford, OH).

47. Eaton DW, Davidsen J, Pedersen PK, Boroumand N (2014) Breakdown of theGutenberg-Richter relation for microearthquakes induced by hydraulic fracturing:Influence of stratabound fractures. Geophys Prospect 62:806–818.

48. Maghsoudi S, Eaton DW, Davidsen J (2016) Nontrivial clustering of microseismicityinduced by hydraulic fracturing. Geophys Res Lett 43:10672–10679.

49. Warpinski NR, Du J, Zimmer U (2012) Measurements of hydraulic-fracture-inducedseismicity in gas shales. SPE Prod Oper 27:240–252.

50. Goebel THW, Kwiatek G, Becker TW, Brodsky EE, Dresen G (2017) What allows seismicevents to grow big?: Insights from b-value and fault roughness analysis in laboratorystick-slip experiments. Geology 45:815–818.

51. Perrin C, Manighetti I, Ampuero JP, Cappa F, Gaudemer Y (2016) Location of largestearthquake slip and fast rupture controlled by along‐strike change in fault structuralmaturity due to fault growth. J Geophys Res Solid Earth 121:3666–3685.

52. Savage HM, Brodsky EE (2011) Collateral damage: Evolution with displacement offracture distribution and secondary fault strands in fault damage zones. J GeophysRes 116:B03405.

53. Ikari MJ, Marone C, Saffer DM (2011) On the relation between fault strength andfrictional stability. Geology 39:83–86.

54. Mori J, Abercrombie RE (1997) Depth dependence of earthquake frequency-magnitude distributions in California: Implications for rupture initiation. J GeophysRes 102:15081–15090.

55. Solis MP (2016) Sinistral transpression along previously existing faults, AppalachianBasin, Ohio. AAPG Eastern Section Meeting 2016 Abstracts (American Association ofPetroleum Geologists, Tulsa, OK), article no. 90258.

56. Clifford MJ (1973) Silurian rock salt in Ohio: Report of Investigations (Ohio Division ofGeological Survey, Columbus, OH), RI 90.

57. Baranoski MT (1993) The Cambridge Monocline: A revisitation of a major positivestructural inversion in southeastern Ohio. Proceedings of the First TechnicalSymposium (Ohio Geological Society, Columbus, OH), pp 1–9.

58. Root S (1996) Recurrent basement faulting and basin evolution, West Virginia andOhio: The Burning Springs–Cambridge fault zone. Basement and Basins of EasternNorth America, eds van der Pluijm BA, Catacosinos PA (Geological Society of America,Boulder, CO), Geological Society of America Special Paper, Vol 308, pp 127–134.

59. Baranoski MT, Riley RA (2013) Analysis of stratigraphic, structural, and productionrelationships of Devonian shale gas reservoirs in Meigs County, Ohio (Ohio De-partment of Natural Resources, Division of Geological Survey, Columbus, OH), OpenFile Report 88-3.

60. Shapiro SA, Patzig R, Rothert E, Rindshwentner J (2003) Triggering of seismicity bypore-pressure perturbations: Permeability-related signatures of the phenomenon.Pure Appl Geophys 160:1051–1066.

61. Holtkamp SG, Brudzinski MR, Currie BS (2015) Remote detection and monitoring ofinjection-induced seismicity: Application to the 2010-12 Youngstown, Ohio seismicsequence. AAPG Bull 99:1671–1688.

62. Schultz R, Wang R, Gu YJ, Haug K, Atkinson G (2017) A seismological overview of theinduced earthquakes in the Duvernay play near Fox Creek, Alberta. J Geophys ResSolid Earth 122:492–505.

63. Bao X, Eaton DW (2016) Fault activation by hydraulic fracturing in western Canada.Science 354:1406–1409.

64. Anderson RN (1980) Phase changes & the frequency-magnitude distribution in theupper plane of the deep seismic zone beneath Tohoku, Japan. J Geophys Res 85:1389–1398.

65. Wiemer S, McNutt SR (1997) Variations in the frequency magnitude distribution withdepth in two volcanic areas: Mount St Helens, Washington, and Mt Spurr, Alaska.Geophys Res Lett 24:189–192.

66. Lei X, Yu G, Ma S, Wen X, Wang Q (2008) Earthquakes induced by water injection at∼3 km depth within the Rongchang gas field, Chongqing, China. J Geophys Res 113:B10310.

67. McGarr A (2014) Maximum magnitude earthquakes induced by fluid injection.J Geophys Res Solid Earth 119:1008–1019.

68. Shapiro SA, Krüger OS, Dinske C, Langenbruch C (2011) Magnitudes of inducedearthquakes and geometric scales of fluid-stimulated rock volume. Geophysics 76:55–63.

69. Dieterich J (1994) A constitutive law for rate of earthquake production and its ap-plication to earthquake clustering. J Geophys Res 99:2601–2618.

70. Doser DI, Kanamori H (1986) Depth of seismicity in the imperial valley region (1977-1983) and its relationship to heat flow, crustal structure, and the October 15, 1979,earthquake. J Geophys Res 91:675–688.

71. Marone C, Scholz CH (1988) The depth of seismic faulting and the upper transitionfrom stable to unstable slip regimes. Geophys Res Lett 15:621–624.

72. Ben-Zion Y, Lyakhovsky V (2006) Analysis of aftershocks in a lithospheric model withseismogenic zone governed by damage rheology. Heophys J Int 165:197–210.

73. Hagin PN, Zoback MD (2004) Viscous deformation of unconsolidated sands–Part 1:Time-dependent deformation, frequency dispersion, and attenuation. Geophysics 69:731–741.

74. Sone H, Zoback MD (2014) Time-dependent deformation of shale gas reservoir rocks andits long-term effect on the in situ state of stress. Int J Rock Mech Min Sci 69:120–132.

E1728 | www.pnas.org/cgi/doi/10.1073/pnas.1715284115 Kozłowska et al.

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

7, 2

020

75. Herrmann RB (2013) Computer programs in seismology: An evolving tool for in-struction & research. Seismol Res Lett 84:1081–1088.

76. Waldhauser F, Ellsworth WL (2000) A double-difference earthquake location algo-rithm: Method & application to the northern Hayward fault, California. Bull SeismolSoc Am 90:1353–1368.

77. Chen W-P, Molnar P (1983) Focal depths of intracontinental and intraplate earth-quakes and their implications for the thermal and mechanical properties of thelithosphere. J Geophys Res 88:4183–4214.

78. Bender B (1983) Maximum likelihood estimation of b values for magnitude groupeddata. Bull Seismol Soc Am 73:831–851.

79. Wiemer S, Wyss M (2000) Minimum magnitude of completeness in earthquake catalogs:Examples fromAlaska, thewestern United States, and Japan. Bull Seismol Soc Am 90:859–869.

80. Snoke JA (2003) FOCMEC: FOCal MEChanism Determinations Package. Version6.28.03. Available at www.iris.edu/pub/programs/focmec/. Accessed February 2015.

81. Brudzinski MR, Chen W-P (2003) Visualization of seismicity along subduction zones:Toward a physical basis. Seismol Res Lett 74:731–738.

Kozłowska et al. PNAS | Published online February 5, 2018 | E1729

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

7, 2

020